Genetic Algorithm in Machine Learning for the Tendency Prediction of Taiwan Stock Exchange Weighted Index

[[abstract]]本研究利用機器學習模型，倒傳遞類神經網路以及隨機森林，以總體經濟金融指標作為輸入特徵，預測台灣加權股價指數未來每月的漲跌，並將預測結果實證在期貨上。此外本研究結合遺傳演算法來優化類神經網路以及隨機森林之參數，且經由遺傳演算法及隨機森林做挑選特徵，觀看上述方法能否提升投資績效。 實證結果顯示，經過特徵挑選的類神經網路，其績效比未經過特徵挑選的類神經網路以及隨機森林都好。其中以遺傳演算法挑選特徵又比以隨機森林挑選特徵之績效更好。預測漲的準確率皆在五成五以上，其中兩個經過挑選特徵之類神經網路更能到六成以上．預測跌的部分除了以遺傳演算法篩選特徵之類神經網路，其餘準確率皆在五成以下，其中以隨機森林篩選特徵之類神經網路雖然其準確率不及五成，但仍可創造正的投資報酬率。 本研究證實以機器學習模型可以應用在投資領域，預測台灣加權股價指數的月趨勢，藉由預測的結果搭配合適的交易策略，發展一套可實際運用且有效的投資系統。[[abstract]]The research model of this model is machine learning model including Back Propagation Neural Networks and Random Forest This study uses macroeconomic data as features to predict the movement of TAIEX in the following month and invests in future according to the results of the prediction Besides this study combine Genetic Algorithm to optimize the parameters in the Neural Networks model and Random Forest model Moreover this study selects features with both Genetic Algorithm and Random Forest In the end this study shows the performance of the investment in the future We expect that the method described above can improve the investment performance Research results are given that Neural Networks model with frature selection have better performance Neural Networks with features selected with Genetic Algorithm has better performance than one with Random Forest Accuracy rates of up prediction are more than 55% besides accuracy rates of two of the Neural Networks combined with feature selection are more than 60% On the other hand accuracy rates of down prediction are less than 50% Although accuracy rates of two of the Neural Networks models combined with feature selection prediction down are less than 50% the models can gain profits This study applies machine learning model on investment field to predict the trend of Taiwan Stock Market Index After finishing prediction we create appropriate strategies depending on the prediction results This study develops a practical and effective investment system with the process described abov